Google's AI research agents discover 66 novel biomarkers in automated biomedical study

imjustnewatai · x · 2026-08-24

Google quietly built a software biomedical research group driven by agents. Starting with wearable data, specialized agents form hypotheses, run statistics, attack weak results, verify biology, and assemble papers. Across 9,279 participant-observations, it prioritized 41 mental-health and 25 metabolic candidate biomarkers and derived a steps/resting-heart-rate index linked to insulin resistance. In a blind study, 15 experts estimated they would retain 56.9% of its manuscript content, compared to 18.8–30.4% for other AI systems. It was the only system to receive "accept" or "minor revision" recommendations. The 66 signals require clinical validation.

Original post →

More from coding & agent

coding & agent channel →